fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/vmacache.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/tlb.h>
  63
  64 #include <trace/events/task.h>
  65 #include "internal.h"
  66
  67 #include <trace/events/sched.h>
  68
  69 int suid_dumpable = 0;
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         BUG_ON(!fmt);
  77         if (WARN_ON(!fmt->load_binary))
  78                 return;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 #ifdef CONFIG_USELIB
 102 /*
 103  * Note that a shared library must be both readable and executable due to
 104  * security reasons.
 105  *
 106  * Also note that we take the address to load from from the file itself.
 107  */
 108 SYSCALL_DEFINE1(uselib, const char __user *, library)
 109 {
 110         struct linux_binfmt *fmt;
 111         struct file *file;
 112         struct filename *tmp = getname(library);
 113         int error = PTR_ERR(tmp);
 114         static const struct open_flags uselib_flags = {
 115                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 116                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 117                 .intent = LOOKUP_OPEN,
 118                 .lookup_flags = LOOKUP_FOLLOW,
 119         };
 120
 121         if (IS_ERR(tmp))
 122                 goto out;
 123
 124         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 125         putname(tmp);
 126         error = PTR_ERR(file);
 127         if (IS_ERR(file))
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(file_inode(file)->i_mode))
 132                 goto exit;
 133
 134         error = -EACCES;
 135         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 136                 goto exit;
 137
 138         fsnotify_open(file);
 139
 140         error = -ENOEXEC;
 141
 142         read_lock(&binfmt_lock);
 143         list_for_each_entry(fmt, &formats, lh) {
 144                 if (!fmt->load_shlib)
 145                         continue;
 146                 if (!try_module_get(fmt->module))
 147                         continue;
 148                 read_unlock(&binfmt_lock);
 149                 error = fmt->load_shlib(file);
 150                 read_lock(&binfmt_lock);
 151                 put_binfmt(fmt);
 152                 if (error != -ENOEXEC)
 153                         break;
 154         }
 155         read_unlock(&binfmt_lock);
 156 exit:
 157         fput(file);
 158 out:
 159         return error;
 160 }
 161 #endif /* #ifdef CONFIG_USELIB */
 162
 163 #ifdef CONFIG_MMU
 164 /*
 165  * The nascent bprm->mm is not visible until exec_mmap() but it can
 166  * use a lot of memory, account these pages in current->mm temporary
 167  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 168  * change the counter back via acct_arg_size(0).
 169  */
 170 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 171 {
 172         struct mm_struct *mm = current->mm;
 173         long diff = (long)(pages - bprm->vma_pages);
 174
 175         if (!mm || !diff)
 176                 return;
 177
 178         bprm->vma_pages = pages;
 179         add_mm_counter(mm, MM_ANONPAGES, diff);
 180 }
 181
 182 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 183                 int write)
 184 {
 185         struct page *page;
 186         int ret;
 187
 188 #ifdef CONFIG_STACK_GROWSUP
 189         if (write) {
 190                 ret = expand_downwards(bprm->vma, pos);
 191                 if (ret < 0)
 192                         return NULL;
 193         }
 194 #endif
 195         ret = get_user_pages(current, bprm->mm, pos,
 196                         1, write, 1, &page, NULL);
 197         if (ret <= 0)
 198                 return NULL;
 199
 200         if (write) {
 201                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 202                 struct rlimit *rlim;
 203
 204                 acct_arg_size(bprm, size / PAGE_SIZE);
 205
 206                 /*
 207                  * We've historically supported up to 32 pages (ARG_MAX)
 208                  * of argument strings even with small stacks
 209                  */
 210                 if (size <= ARG_MAX)
 211                         return page;
 212
 213                 /*
 214                  * Limit to 1/4-th the stack size for the argv+env strings.
 215                  * This ensures that:
 216                  *  - the remaining binfmt code will not run out of stack space,
 217                  *  - the program will have a reasonable amount of stack left
 218                  *    to work from.
 219                  */
 220                 rlim = current->signal->rlim;
 221                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 222                         put_page(page);
 223                         return NULL;
 224                 }
 225         }
 226
 227         return page;
 228 }
 229
 230 static void put_arg_page(struct page *page)
 231 {
 232         put_page(page);
 233 }
 234
 235 static void free_arg_page(struct linux_binprm *bprm, int i)
 236 {
 237 }
 238
 239 static void free_arg_pages(struct linux_binprm *bprm)
 240 {
 241 }
 242
 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 244                 struct page *page)
 245 {
 246         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 247 }
 248
 249 static int __bprm_mm_init(struct linux_binprm *bprm)
 250 {
 251         int err;
 252         struct vm_area_struct *vma = NULL;
 253         struct mm_struct *mm = bprm->mm;
 254
 255         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 256         if (!vma)
 257                 return -ENOMEM;
 258
 259         down_write(&mm->mmap_sem);
 260         vma->vm_mm = mm;
 261
 262         /*
 263          * Place the stack at the largest stack address the architecture
 264          * supports. Later, we'll move this to an appropriate place. We don't
 265          * use STACK_TOP because that can depend on attributes which aren't
 266          * configured yet.
 267          */
 268         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 269         vma->vm_end = STACK_TOP_MAX;
 270         vma->vm_start = vma->vm_end - PAGE_SIZE;
 271         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 272         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 273         INIT_LIST_HEAD(&vma->anon_vma_chain);
 274
 275         err = insert_vm_struct(mm, vma);
 276         if (err)
 277                 goto err;
 278
 279         mm->stack_vm = mm->total_vm = 1;
 280         arch_bprm_mm_init(mm, vma);
 281         up_write(&mm->mmap_sem);
 282         bprm->p = vma->vm_end - sizeof(void *);
 283         return 0;
 284 err:
 285         up_write(&mm->mmap_sem);
 286         bprm->vma = NULL;
 287         kmem_cache_free(vm_area_cachep, vma);
 288         return err;
 289 }
 290
 291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 292 {
 293         return len <= MAX_ARG_STRLEN;
 294 }
 295
 296 #else
 297
 298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 299 {
 300 }
 301
 302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 303                 int write)
 304 {
 305         struct page *page;
 306
 307         page = bprm->page[pos / PAGE_SIZE];
 308         if (!page && write) {
 309                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 310                 if (!page)
 311                         return NULL;
 312                 bprm->page[pos / PAGE_SIZE] = page;
 313         }
 314
 315         return page;
 316 }
 317
 318 static void put_arg_page(struct page *page)
 319 {
 320 }
 321
 322 static void free_arg_page(struct linux_binprm *bprm, int i)
 323 {
 324         if (bprm->page[i]) {
 325                 __free_page(bprm->page[i]);
 326                 bprm->page[i] = NULL;
 327         }
 328 }
 329
 330 static void free_arg_pages(struct linux_binprm *bprm)
 331 {
 332         int i;
 333
 334         for (i = 0; i < MAX_ARG_PAGES; i++)
 335                 free_arg_page(bprm, i);
 336 }
 337
 338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 339                 struct page *page)
 340 {
 341 }
 342
 343 static int __bprm_mm_init(struct linux_binprm *bprm)
 344 {
 345         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 346         return 0;
 347 }
 348
 349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 350 {
 351         return len <= bprm->p;
 352 }
 353
 354 #endif /* CONFIG_MMU */
 355
 356 /*
 357  * Create a new mm_struct and populate it with a temporary stack
 358  * vm_area_struct.  We don't have enough context at this point to set the stack
 359  * flags, permissions, and offset, so we use temporary values.  We'll update
 360  * them later in setup_arg_pages().
 361  */
 362 static int bprm_mm_init(struct linux_binprm *bprm)
 363 {
 364         int err;
 365         struct mm_struct *mm = NULL;
 366
 367         bprm->mm = mm = mm_alloc();
 368         err = -ENOMEM;
 369         if (!mm)
 370                 goto err;
 371
 372         err = __bprm_mm_init(bprm);
 373         if (err)
 374                 goto err;
 375
 376         return 0;
 377
 378 err:
 379         if (mm) {
 380                 bprm->mm = NULL;
 381                 mmdrop(mm);
 382         }
 383
 384         return err;
 385 }
 386
 387 struct user_arg_ptr {
 388 #ifdef CONFIG_COMPAT
 389         bool is_compat;
 390 #endif
 391         union {
 392                 const char __user *const __user *native;
 393 #ifdef CONFIG_COMPAT
 394                 const compat_uptr_t __user *compat;
 395 #endif
 396         } ptr;
 397 };
 398
 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 400 {
 401         const char __user *native;
 402
 403 #ifdef CONFIG_COMPAT
 404         if (unlikely(argv.is_compat)) {
 405                 compat_uptr_t compat;
 406
 407                 if (get_user(compat, argv.ptr.compat + nr))
 408                         return ERR_PTR(-EFAULT);
 409
 410                 return compat_ptr(compat);
 411         }
 412 #endif
 413
 414         if (get_user(native, argv.ptr.native + nr))
 415                 return ERR_PTR(-EFAULT);
 416
 417         return native;
 418 }
 419
 420 /*
 421  * count() counts the number of strings in array ARGV.
 422  */
 423 static int count(struct user_arg_ptr argv, int max)
 424 {
 425         int i = 0;
 426
 427         if (argv.ptr.native != NULL) {
 428                 for (;;) {
 429                         const char __user *p = get_user_arg_ptr(argv, i);
 430
 431                         if (!p)
 432                                 break;
 433
 434                         if (IS_ERR(p))
 435                                 return -EFAULT;
 436
 437                         if (i >= max)
 438                                 return -E2BIG;
 439                         ++i;
 440
 441                         if (fatal_signal_pending(current))
 442                                 return -ERESTARTNOHAND;
 443                         cond_resched();
 444                 }
 445         }
 446         return i;
 447 }
 448
 449 /*
 450  * 'copy_strings()' copies argument/environment strings from the old
 451  * processes's memory to the new process's stack.  The call to get_user_pages()
 452  * ensures the destination page is created and not swapped out.
 453  */
 454 static int copy_strings(int argc, struct user_arg_ptr argv,
 455                         struct linux_binprm *bprm)
 456 {
 457         struct page *kmapped_page = NULL;
 458         char *kaddr = NULL;
 459         unsigned long kpos = 0;
 460         int ret;
 461
 462         while (argc-- > 0) {
 463                 const char __user *str;
 464                 int len;
 465                 unsigned long pos;
 466
 467                 ret = -EFAULT;
 468                 str = get_user_arg_ptr(argv, argc);
 469                 if (IS_ERR(str))
 470                         goto out;
 471
 472                 len = strnlen_user(str, MAX_ARG_STRLEN);
 473                 if (!len)
 474                         goto out;
 475
 476                 ret = -E2BIG;
 477                 if (!valid_arg_len(bprm, len))
 478                         goto out;
 479
 480                 /* We're going to work our way backwords. */
 481                 pos = bprm->p;
 482                 str += len;
 483                 bprm->p -= len;
 484
 485                 while (len > 0) {
 486                         int offset, bytes_to_copy;
 487
 488                         if (fatal_signal_pending(current)) {
 489                                 ret = -ERESTARTNOHAND;
 490                                 goto out;
 491                         }
 492                         cond_resched();
 493
 494                         offset = pos % PAGE_SIZE;
 495                         if (offset == 0)
 496                                 offset = PAGE_SIZE;
 497
 498                         bytes_to_copy = offset;
 499                         if (bytes_to_copy > len)
 500                                 bytes_to_copy = len;
 501
 502                         offset -= bytes_to_copy;
 503                         pos -= bytes_to_copy;
 504                         str -= bytes_to_copy;
 505                         len -= bytes_to_copy;
 506
 507                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 508                                 struct page *page;
 509
 510                                 page = get_arg_page(bprm, pos, 1);
 511                                 if (!page) {
 512                                         ret = -E2BIG;
 513                                         goto out;
 514                                 }
 515
 516                                 if (kmapped_page) {
 517                                         flush_kernel_dcache_page(kmapped_page);
 518                                         kunmap(kmapped_page);
 519                                         put_arg_page(kmapped_page);
 520                                 }
 521                                 kmapped_page = page;
 522                                 kaddr = kmap(kmapped_page);
 523                                 kpos = pos & PAGE_MASK;
 524                                 flush_arg_page(bprm, kpos, kmapped_page);
 525                         }
 526                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 527                                 ret = -EFAULT;
 528                                 goto out;
 529                         }
 530                 }
 531         }
 532         ret = 0;
 533 out:
 534         if (kmapped_page) {
 535                 flush_kernel_dcache_page(kmapped_page);
 536                 kunmap(kmapped_page);
 537                 put_arg_page(kmapped_page);
 538         }
 539         return ret;
 540 }
 541
 542 /*
 543  * Like copy_strings, but get argv and its values from kernel memory.
 544  */
 545 int copy_strings_kernel(int argc, const char *const *__argv,
 546                         struct linux_binprm *bprm)
 547 {
 548         int r;
 549         mm_segment_t oldfs = get_fs();
 550         struct user_arg_ptr argv = {
 551                 .ptr.native = (const char __user *const  __user *)__argv,
 552         };
 553
 554         set_fs(KERNEL_DS);
 555         r = copy_strings(argc, argv, bprm);
 556         set_fs(oldfs);
 557
 558         return r;
 559 }
 560 EXPORT_SYMBOL(copy_strings_kernel);
 561
 562 #ifdef CONFIG_MMU
 563
 564 /*
 565  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 566  * the binfmt code determines where the new stack should reside, we shift it to
 567  * its final location.  The process proceeds as follows:
 568  *
 569  * 1) Use shift to calculate the new vma endpoints.
 570  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 571  *    arguments passed to subsequent functions are consistent.
 572  * 3) Move vma's page tables to the new range.
 573  * 4) Free up any cleared pgd range.
 574  * 5) Shrink the vma to cover only the new range.
 575  */
 576 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 577 {
 578         struct mm_struct *mm = vma->vm_mm;
 579         unsigned long old_start = vma->vm_start;
 580         unsigned long old_end = vma->vm_end;
 581         unsigned long length = old_end - old_start;
 582         unsigned long new_start = old_start - shift;
 583         unsigned long new_end = old_end - shift;
 584         struct mmu_gather tlb;
 585
 586         BUG_ON(new_start > new_end);
 587
 588         /*
 589          * ensure there are no vmas between where we want to go
 590          * and where we are
 591          */
 592         if (vma != find_vma(mm, new_start))
 593                 return -EFAULT;
 594
 595         /*
 596          * cover the whole range: [new_start, old_end)
 597          */
 598         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 599                 return -ENOMEM;
 600
 601         /*
 602          * move the page tables downwards, on failure we rely on
 603          * process cleanup to remove whatever mess we made.
 604          */
 605         if (length != move_page_tables(vma, old_start,
 606                                        vma, new_start, length, false))
 607                 return -ENOMEM;
 608
 609         lru_add_drain();
 610         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 611         if (new_end > old_start) {
 612                 /*
 613                  * when the old and new regions overlap clear from new_end.
 614                  */
 615                 free_pgd_range(&tlb, new_end, old_end, new_end,
 616                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 617         } else {
 618                 /*
 619                  * otherwise, clean from old_start; this is done to not touch
 620                  * the address space in [new_end, old_start) some architectures
 621                  * have constraints on va-space that make this illegal (IA64) -
 622                  * for the others its just a little faster.
 623                  */
 624                 free_pgd_range(&tlb, old_start, old_end, new_end,
 625                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 626         }
 627         tlb_finish_mmu(&tlb, old_start, old_end);
 628
 629         /*
 630          * Shrink the vma to just the new range.  Always succeeds.
 631          */
 632         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 633
 634         return 0;
 635 }
 636
 637 /*
 638  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 639  * the stack is optionally relocated, and some extra space is added.
 640  */
 641 int setup_arg_pages(struct linux_binprm *bprm,
 642                     unsigned long stack_top,
 643                     int executable_stack)
 644 {
 645         unsigned long ret;
 646         unsigned long stack_shift;
 647         struct mm_struct *mm = current->mm;
 648         struct vm_area_struct *vma = bprm->vma;
 649         struct vm_area_struct *prev = NULL;
 650         unsigned long vm_flags;
 651         unsigned long stack_base;
 652         unsigned long stack_size;
 653         unsigned long stack_expand;
 654         unsigned long rlim_stack;
 655
 656 #ifdef CONFIG_STACK_GROWSUP
 657         /* Limit stack size */
 658         stack_base = rlimit_max(RLIMIT_STACK);
 659         if (stack_base > STACK_SIZE_MAX)
 660                 stack_base = STACK_SIZE_MAX;
 661
 662         /* Make sure we didn't let the argument array grow too large. */
 663         if (vma->vm_end - vma->vm_start > stack_base)
 664                 return -ENOMEM;
 665
 666         stack_base = PAGE_ALIGN(stack_top - stack_base);
 667
 668         stack_shift = vma->vm_start - stack_base;
 669         mm->arg_start = bprm->p - stack_shift;
 670         bprm->p = vma->vm_end - stack_shift;
 671 #else
 672         stack_top = arch_align_stack(stack_top);
 673         stack_top = PAGE_ALIGN(stack_top);
 674
 675         if (unlikely(stack_top < mmap_min_addr) ||
 676             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 677                 return -ENOMEM;
 678
 679         stack_shift = vma->vm_end - stack_top;
 680
 681         bprm->p -= stack_shift;
 682         mm->arg_start = bprm->p;
 683 #endif
 684
 685         if (bprm->loader)
 686                 bprm->loader -= stack_shift;
 687         bprm->exec -= stack_shift;
 688
 689         down_write(&mm->mmap_sem);
 690         vm_flags = VM_STACK_FLAGS;
 691
 692         /*
 693          * Adjust stack execute permissions; explicitly enable for
 694          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 695          * (arch default) otherwise.
 696          */
 697         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 698                 vm_flags |= VM_EXEC;
 699         else if (executable_stack == EXSTACK_DISABLE_X)
 700                 vm_flags &= ~VM_EXEC;
 701         vm_flags |= mm->def_flags;
 702         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 703
 704         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 705                         vm_flags);
 706         if (ret)
 707                 goto out_unlock;
 708         BUG_ON(prev != vma);
 709
 710         /* Move stack pages down in memory. */
 711         if (stack_shift) {
 712                 ret = shift_arg_pages(vma, stack_shift);
 713                 if (ret)
 714                         goto out_unlock;
 715         }
 716
 717         /* mprotect_fixup is overkill to remove the temporary stack flags */
 718         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 719
 720         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 721         stack_size = vma->vm_end - vma->vm_start;
 722         /*
 723          * Align this down to a page boundary as expand_stack
 724          * will align it up.
 725          */
 726         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 727 #ifdef CONFIG_STACK_GROWSUP
 728         if (stack_size + stack_expand > rlim_stack)
 729                 stack_base = vma->vm_start + rlim_stack;
 730         else
 731                 stack_base = vma->vm_end + stack_expand;
 732 #else
 733         if (stack_size + stack_expand > rlim_stack)
 734                 stack_base = vma->vm_end - rlim_stack;
 735         else
 736                 stack_base = vma->vm_start - stack_expand;
 737 #endif
 738         current->mm->start_stack = bprm->p;
 739         ret = expand_stack(vma, stack_base);
 740         if (ret)
 741                 ret = -EFAULT;
 742
 743 out_unlock:
 744         up_write(&mm->mmap_sem);
 745         return ret;
 746 }
 747 EXPORT_SYMBOL(setup_arg_pages);
 748
 749 #endif /* CONFIG_MMU */
 750
 751 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 752 {
 753         struct file *file;
 754         int err;
 755         struct open_flags open_exec_flags = {
 756                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 757                 .acc_mode = MAY_EXEC | MAY_OPEN,
 758                 .intent = LOOKUP_OPEN,
 759                 .lookup_flags = LOOKUP_FOLLOW,
 760         };
 761
 762         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 763                 return ERR_PTR(-EINVAL);
 764         if (flags & AT_SYMLINK_NOFOLLOW)
 765                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 766         if (flags & AT_EMPTY_PATH)
 767                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 768
 769         file = do_filp_open(fd, name, &open_exec_flags);
 770         if (IS_ERR(file))
 771                 goto out;
 772
 773         err = -EACCES;
 774         if (!S_ISREG(file_inode(file)->i_mode))
 775                 goto exit;
 776
 777         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 778                 goto exit;
 779
 780         err = deny_write_access(file);
 781         if (err)
 782                 goto exit;
 783
 784         if (name->name[0] != '\0')
 785                 fsnotify_open(file);
 786
 787 out:
 788         return file;
 789
 790 exit:
 791         fput(file);
 792         return ERR_PTR(err);
 793 }
 794
 795 struct file *open_exec(const char *name)
 796 {
 797         struct filename *filename = getname_kernel(name);
 798         struct file *f = ERR_CAST(filename);
 799
 800         if (!IS_ERR(filename)) {
 801                 f = do_open_execat(AT_FDCWD, filename, 0);
 802                 putname(filename);
 803         }
 804         return f;
 805 }
 806 EXPORT_SYMBOL(open_exec);
 807
 808 int kernel_read(struct file *file, loff_t offset,
 809                 char *addr, unsigned long count)
 810 {
 811         mm_segment_t old_fs;
 812         loff_t pos = offset;
 813         int result;
 814
 815         old_fs = get_fs();
 816         set_fs(get_ds());
 817         /* The cast to a user pointer is valid due to the set_fs() */
 818         result = vfs_read(file, (void __user *)addr, count, &pos);
 819         set_fs(old_fs);
 820         return result;
 821 }
 822
 823 EXPORT_SYMBOL(kernel_read);
 824
 825 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 826 {
 827         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 828         if (res > 0)
 829                 flush_icache_range(addr, addr + len);
 830         return res;
 831 }
 832 EXPORT_SYMBOL(read_code);
 833
 834 static int exec_mmap(struct mm_struct *mm)
 835 {
 836         struct task_struct *tsk;
 837         struct mm_struct *old_mm, *active_mm;
 838
 839         /* Notify parent that we're no longer interested in the old VM */
 840         tsk = current;
 841         old_mm = current->mm;
 842         mm_release(tsk, old_mm);
 843
 844         if (old_mm) {
 845                 sync_mm_rss(old_mm);
 846                 /*
 847                  * Make sure that if there is a core dump in progress
 848                  * for the old mm, we get out and die instead of going
 849                  * through with the exec.  We must hold mmap_sem around
 850                  * checking core_state and changing tsk->mm.
 851                  */
 852                 down_read(&old_mm->mmap_sem);
 853                 if (unlikely(old_mm->core_state)) {
 854                         up_read(&old_mm->mmap_sem);
 855                         return -EINTR;
 856                 }
 857         }
 858         task_lock(tsk);
 859         active_mm = tsk->active_mm;
 860         tsk->mm = mm;
 861         tsk->active_mm = mm;
 862         activate_mm(active_mm, mm);
 863         tsk->mm->vmacache_seqnum = 0;
 864         vmacache_flush(tsk);
 865         task_unlock(tsk);
 866         if (old_mm) {
 867                 up_read(&old_mm->mmap_sem);
 868                 BUG_ON(active_mm != old_mm);
 869                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 870                 mm_update_next_owner(old_mm);
 871                 mmput(old_mm);
 872                 return 0;
 873         }
 874         mmdrop(active_mm);
 875         return 0;
 876 }
 877
 878 /*
 879  * This function makes sure the current process has its own signal table,
 880  * so that flush_signal_handlers can later reset the handlers without
 881  * disturbing other processes.  (Other processes might share the signal
 882  * table via the CLONE_SIGHAND option to clone().)
 883  */
 884 static int de_thread(struct task_struct *tsk)
 885 {
 886         struct signal_struct *sig = tsk->signal;
 887         struct sighand_struct *oldsighand = tsk->sighand;
 888         spinlock_t *lock = &oldsighand->siglock;
 889
 890         if (thread_group_empty(tsk))
 891                 goto no_thread_group;
 892
 893         /*
 894          * Kill all other threads in the thread group.
 895          */
 896         spin_lock_irq(lock);
 897         if (signal_group_exit(sig)) {
 898                 /*
 899                  * Another group action in progress, just
 900                  * return so that the signal is processed.
 901                  */
 902                 spin_unlock_irq(lock);
 903                 return -EAGAIN;
 904         }
 905
 906         sig->group_exit_task = tsk;
 907         sig->notify_count = zap_other_threads(tsk);
 908         if (!thread_group_leader(tsk))
 909                 sig->notify_count--;
 910
 911         while (sig->notify_count) {
 912                 __set_current_state(TASK_KILLABLE);
 913                 spin_unlock_irq(lock);
 914                 schedule();
 915                 if (unlikely(__fatal_signal_pending(tsk)))
 916                         goto killed;
 917                 spin_lock_irq(lock);
 918         }
 919         spin_unlock_irq(lock);
 920
 921         /*
 922          * At this point all other threads have exited, all we have to
 923          * do is to wait for the thread group leader to become inactive,
 924          * and to assume its PID:
 925          */
 926         if (!thread_group_leader(tsk)) {
 927                 struct task_struct *leader = tsk->group_leader;
 928
 929                 sig->notify_count = -1; /* for exit_notify() */
 930                 for (;;) {
 931                         threadgroup_change_begin(tsk);
 932                         write_lock_irq(&tasklist_lock);
 933                         if (likely(leader->exit_state))
 934                                 break;
 935                         __set_current_state(TASK_KILLABLE);
 936                         write_unlock_irq(&tasklist_lock);
 937                         threadgroup_change_end(tsk);
 938                         schedule();
 939                         if (unlikely(__fatal_signal_pending(tsk)))
 940                                 goto killed;
 941                 }
 942
 943                 /*
 944                  * The only record we have of the real-time age of a
 945                  * process, regardless of execs it's done, is start_time.
 946                  * All the past CPU time is accumulated in signal_struct
 947                  * from sister threads now dead.  But in this non-leader
 948                  * exec, nothing survives from the original leader thread,
 949                  * whose birth marks the true age of this process now.
 950                  * When we take on its identity by switching to its PID, we
 951                  * also take its birthdate (always earlier than our own).
 952                  */
 953                 tsk->start_time = leader->start_time;
 954                 tsk->real_start_time = leader->real_start_time;
 955
 956                 BUG_ON(!same_thread_group(leader, tsk));
 957                 BUG_ON(has_group_leader_pid(tsk));
 958                 /*
 959                  * An exec() starts a new thread group with the
 960                  * TGID of the previous thread group. Rehash the
 961                  * two threads with a switched PID, and release
 962                  * the former thread group leader:
 963                  */
 964
 965                 /* Become a process group leader with the old leader's pid.
 966                  * The old leader becomes a thread of the this thread group.
 967                  * Note: The old leader also uses this pid until release_task
 968                  *       is called.  Odd but simple and correct.
 969                  */
 970                 tsk->pid = leader->pid;
 971                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 972                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 973                 transfer_pid(leader, tsk, PIDTYPE_SID);
 974
 975                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 976                 list_replace_init(&leader->sibling, &tsk->sibling);
 977
 978                 tsk->group_leader = tsk;
 979                 leader->group_leader = tsk;
 980
 981                 tsk->exit_signal = SIGCHLD;
 982                 leader->exit_signal = -1;
 983
 984                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 985                 leader->exit_state = EXIT_DEAD;
 986
 987                 /*
 988                  * We are going to release_task()->ptrace_unlink() silently,
 989                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 990                  * the tracer wont't block again waiting for this thread.
 991                  */
 992                 if (unlikely(leader->ptrace))
 993                         __wake_up_parent(leader, leader->parent);
 994                 write_unlock_irq(&tasklist_lock);
 995                 threadgroup_change_end(tsk);
 996
 997                 release_task(leader);
 998         }
 999
1000         sig->group_exit_task = NULL;
1001         sig->notify_count = 0;
1002
1003 no_thread_group:
1004         /* we have changed execution domain */
1005         tsk->exit_signal = SIGCHLD;
1006
1007         exit_itimers(sig);
1008         flush_itimer_signals();
1009
1010         if (atomic_read(&oldsighand->count) != 1) {
1011                 struct sighand_struct *newsighand;
1012                 /*
1013                  * This ->sighand is shared with the CLONE_SIGHAND
1014                  * but not CLONE_THREAD task, switch to the new one.
1015                  */
1016                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1017                 if (!newsighand)
1018                         return -ENOMEM;
1019
1020                 atomic_set(&newsighand->count, 1);
1021                 memcpy(newsighand->action, oldsighand->action,
1022                        sizeof(newsighand->action));
1023
1024                 write_lock_irq(&tasklist_lock);
1025                 spin_lock(&oldsighand->siglock);
1026                 rcu_assign_pointer(tsk->sighand, newsighand);
1027                 spin_unlock(&oldsighand->siglock);
1028                 write_unlock_irq(&tasklist_lock);
1029
1030                 __cleanup_sighand(oldsighand);
1031         }
1032
1033         BUG_ON(!thread_group_leader(tsk));
1034         return 0;
1035
1036 killed:
1037         /* protects against exit_notify() and __exit_signal() */
1038         read_lock(&tasklist_lock);
1039         sig->group_exit_task = NULL;
1040         sig->notify_count = 0;
1041         read_unlock(&tasklist_lock);
1042         return -EAGAIN;
1043 }
1044
1045 char *get_task_comm(char *buf, struct task_struct *tsk)
1046 {
1047         /* buf must be at least sizeof(tsk->comm) in size */
1048         task_lock(tsk);
1049         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1050         task_unlock(tsk);
1051         return buf;
1052 }
1053 EXPORT_SYMBOL_GPL(get_task_comm);
1054
1055 /*
1056  * These functions flushes out all traces of the currently running executable
1057  * so that a new one can be started
1058  */
1059
1060 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1061 {
1062         task_lock(tsk);
1063         trace_task_rename(tsk, buf);
1064         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1065         task_unlock(tsk);
1066         perf_event_comm(tsk, exec);
1067 }
1068
1069 int flush_old_exec(struct linux_binprm * bprm)
1070 {
1071         int retval;
1072
1073         /*
1074          * Make sure we have a private signal table and that
1075          * we are unassociated from the previous thread group.
1076          */
1077         retval = de_thread(current);
1078         if (retval)
1079                 goto out;
1080
1081         set_mm_exe_file(bprm->mm, bprm->file);
1082         /*
1083          * Release all of the old mmap stuff
1084          */
1085         acct_arg_size(bprm, 0);
1086         retval = exec_mmap(bprm->mm);
1087         if (retval)
1088                 goto out;
1089
1090         bprm->mm = NULL;                /* We're using it now */
1091
1092         set_fs(USER_DS);
1093         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1094                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1095         flush_thread();
1096         current->personality &= ~bprm->per_clear;
1097
1098         return 0;
1099
1100 out:
1101         return retval;
1102 }
1103 EXPORT_SYMBOL(flush_old_exec);
1104
1105 void would_dump(struct linux_binprm *bprm, struct file *file)
1106 {
1107         if (inode_permission(file_inode(file), MAY_READ) < 0)
1108                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1109 }
1110 EXPORT_SYMBOL(would_dump);
1111
1112 void setup_new_exec(struct linux_binprm * bprm)
1113 {
1114         arch_pick_mmap_layout(current->mm);
1115
1116         /* This is the point of no return */
1117         current->sas_ss_sp = current->sas_ss_size = 0;
1118
1119         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1120                 set_dumpable(current->mm, SUID_DUMP_USER);
1121         else
1122                 set_dumpable(current->mm, suid_dumpable);
1123
1124         perf_event_exec();
1125         __set_task_comm(current, kbasename(bprm->filename), true);
1126
1127         /* Set the new mm task size. We have to do that late because it may
1128          * depend on TIF_32BIT which is only updated in flush_thread() on
1129          * some architectures like powerpc
1130          */
1131         current->mm->task_size = TASK_SIZE;
1132
1133         /* install the new credentials */
1134         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1135             !gid_eq(bprm->cred->gid, current_egid())) {
1136                 current->pdeath_signal = 0;
1137         } else {
1138                 would_dump(bprm, bprm->file);
1139                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1140                         set_dumpable(current->mm, suid_dumpable);
1141         }
1142
1143         /* An exec changes our domain. We are no longer part of the thread
1144            group */
1145         current->self_exec_id++;
1146         flush_signal_handlers(current, 0);
1147         do_close_on_exec(current->files);
1148 }
1149 EXPORT_SYMBOL(setup_new_exec);
1150
1151 /*
1152  * Prepare credentials and lock ->cred_guard_mutex.
1153  * install_exec_creds() commits the new creds and drops the lock.
1154  * Or, if exec fails before, free_bprm() should release ->cred and
1155  * and unlock.
1156  */
1157 int prepare_bprm_creds(struct linux_binprm *bprm)
1158 {
1159         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1160                 return -ERESTARTNOINTR;
1161
1162         bprm->cred = prepare_exec_creds();
1163         if (likely(bprm->cred))
1164                 return 0;
1165
1166         mutex_unlock(&current->signal->cred_guard_mutex);
1167         return -ENOMEM;
1168 }
1169
1170 static void free_bprm(struct linux_binprm *bprm)
1171 {
1172         free_arg_pages(bprm);
1173         if (bprm->cred) {
1174                 mutex_unlock(&current->signal->cred_guard_mutex);
1175                 abort_creds(bprm->cred);
1176         }
1177         if (bprm->file) {
1178                 allow_write_access(bprm->file);
1179                 fput(bprm->file);
1180         }
1181         /* If a binfmt changed the interp, free it. */
1182         if (bprm->interp != bprm->filename)
1183                 kfree(bprm->interp);
1184         kfree(bprm);
1185 }
1186
1187 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1188 {
1189         /* If a binfmt changed the interp, free it first. */
1190         if (bprm->interp != bprm->filename)
1191                 kfree(bprm->interp);
1192         bprm->interp = kstrdup(interp, GFP_KERNEL);
1193         if (!bprm->interp)
1194                 return -ENOMEM;
1195         return 0;
1196 }
1197 EXPORT_SYMBOL(bprm_change_interp);
1198
1199 /*
1200  * install the new credentials for this executable
1201  */
1202 void install_exec_creds(struct linux_binprm *bprm)
1203 {
1204         security_bprm_committing_creds(bprm);
1205
1206         commit_creds(bprm->cred);
1207         bprm->cred = NULL;
1208
1209         /*
1210          * Disable monitoring for regular users
1211          * when executing setuid binaries. Must
1212          * wait until new credentials are committed
1213          * by commit_creds() above
1214          */
1215         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1216                 perf_event_exit_task(current);
1217         /*
1218          * cred_guard_mutex must be held at least to this point to prevent
1219          * ptrace_attach() from altering our determination of the task's
1220          * credentials; any time after this it may be unlocked.
1221          */
1222         security_bprm_committed_creds(bprm);
1223         mutex_unlock(&current->signal->cred_guard_mutex);
1224 }
1225 EXPORT_SYMBOL(install_exec_creds);
1226
1227 /*
1228  * determine how safe it is to execute the proposed program
1229  * - the caller must hold ->cred_guard_mutex to protect against
1230  *   PTRACE_ATTACH or seccomp thread-sync
1231  */
1232 static void check_unsafe_exec(struct linux_binprm *bprm)
1233 {
1234         struct task_struct *p = current, *t;
1235         unsigned n_fs;
1236
1237         if (p->ptrace) {
1238                 if (p->ptrace & PT_PTRACE_CAP)
1239                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1240                 else
1241                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1242         }
1243
1244         /*
1245          * This isn't strictly necessary, but it makes it harder for LSMs to
1246          * mess up.
1247          */
1248         if (task_no_new_privs(current))
1249                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1250
1251         t = p;
1252         n_fs = 1;
1253         spin_lock(&p->fs->lock);
1254         rcu_read_lock();
1255         while_each_thread(p, t) {
1256                 if (t->fs == p->fs)
1257                         n_fs++;
1258         }
1259         rcu_read_unlock();
1260
1261         if (p->fs->users > n_fs)
1262                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1263         else
1264                 p->fs->in_exec = 1;
1265         spin_unlock(&p->fs->lock);
1266 }
1267
1268 /*
1269  * Fill the binprm structure from the inode.
1270  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1271  *
1272  * This may be called multiple times for binary chains (scripts for example).
1273  */
1274 int prepare_binprm(struct linux_binprm *bprm)
1275 {
1276         struct inode *inode = file_inode(bprm->file);
1277         umode_t mode = inode->i_mode;
1278         int retval;
1279
1280
1281         /* clear any previous set[ug]id data from a previous binary */
1282         bprm->cred->euid = current_euid();
1283         bprm->cred->egid = current_egid();
1284
1285         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1286             !task_no_new_privs(current) &&
1287             kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1288             kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1289                 /* Set-uid? */
1290                 if (mode & S_ISUID) {
1291                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1292                         bprm->cred->euid = inode->i_uid;
1293                 }
1294
1295                 /* Set-gid? */
1296                 /*
1297                  * If setgid is set but no group execute bit then this
1298                  * is a candidate for mandatory locking, not a setgid
1299                  * executable.
1300                  */
1301                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1302                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1303                         bprm->cred->egid = inode->i_gid;
1304                 }
1305         }
1306
1307         /* fill in binprm security blob */
1308         retval = security_bprm_set_creds(bprm);
1309         if (retval)
1310                 return retval;
1311         bprm->cred_prepared = 1;
1312
1313         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1314         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1315 }
1316
1317 EXPORT_SYMBOL(prepare_binprm);
1318
1319 /*
1320  * Arguments are '\0' separated strings found at the location bprm->p
1321  * points to; chop off the first by relocating brpm->p to right after
1322  * the first '\0' encountered.
1323  */
1324 int remove_arg_zero(struct linux_binprm *bprm)
1325 {
1326         int ret = 0;
1327         unsigned long offset;
1328         char *kaddr;
1329         struct page *page;
1330
1331         if (!bprm->argc)
1332                 return 0;
1333
1334         do {
1335                 offset = bprm->p & ~PAGE_MASK;
1336                 page = get_arg_page(bprm, bprm->p, 0);
1337                 if (!page) {
1338                         ret = -EFAULT;
1339                         goto out;
1340                 }
1341                 kaddr = kmap_atomic(page);
1342
1343                 for (; offset < PAGE_SIZE && kaddr[offset];
1344                                 offset++, bprm->p++)
1345                         ;
1346
1347                 kunmap_atomic(kaddr);
1348                 put_arg_page(page);
1349
1350                 if (offset == PAGE_SIZE)
1351                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1352         } while (offset == PAGE_SIZE);
1353
1354         bprm->p++;
1355         bprm->argc--;
1356         ret = 0;
1357
1358 out:
1359         return ret;
1360 }
1361 EXPORT_SYMBOL(remove_arg_zero);
1362
1363 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1364 /*
1365  * cycle the list of binary formats handler, until one recognizes the image
1366  */
1367 int search_binary_handler(struct linux_binprm *bprm)
1368 {
1369         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1370         struct linux_binfmt *fmt;
1371         int retval;
1372
1373         /* This allows 4 levels of binfmt rewrites before failing hard. */
1374         if (bprm->recursion_depth > 5)
1375                 return -ELOOP;
1376
1377         retval = security_bprm_check(bprm);
1378         if (retval)
1379                 return retval;
1380
1381         retval = -ENOENT;
1382  retry:
1383         read_lock(&binfmt_lock);
1384         list_for_each_entry(fmt, &formats, lh) {
1385                 if (!try_module_get(fmt->module))
1386                         continue;
1387                 read_unlock(&binfmt_lock);
1388                 bprm->recursion_depth++;
1389                 retval = fmt->load_binary(bprm);
1390                 read_lock(&binfmt_lock);
1391                 put_binfmt(fmt);
1392                 bprm->recursion_depth--;
1393                 if (retval < 0 && !bprm->mm) {
1394                         /* we got to flush_old_exec() and failed after it */
1395                         read_unlock(&binfmt_lock);
1396                         force_sigsegv(SIGSEGV, current);
1397                         return retval;
1398                 }
1399                 if (retval != -ENOEXEC || !bprm->file) {
1400                         read_unlock(&binfmt_lock);
1401                         return retval;
1402                 }
1403         }
1404         read_unlock(&binfmt_lock);
1405
1406         if (need_retry) {
1407                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1408                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1409                         return retval;
1410                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1411                         return retval;
1412                 need_retry = false;
1413                 goto retry;
1414         }
1415
1416         return retval;
1417 }
1418 EXPORT_SYMBOL(search_binary_handler);
1419
1420 static int exec_binprm(struct linux_binprm *bprm)
1421 {
1422         pid_t old_pid, old_vpid;
1423         int ret;
1424
1425         /* Need to fetch pid before load_binary changes it */
1426         old_pid = current->pid;
1427         rcu_read_lock();
1428         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1429         rcu_read_unlock();
1430
1431         ret = search_binary_handler(bprm);
1432         if (ret >= 0) {
1433                 audit_bprm(bprm);
1434                 trace_sched_process_exec(current, old_pid, bprm);
1435                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1436                 proc_exec_connector(current);
1437         }
1438
1439         return ret;
1440 }
1441
1442 /*
1443  * sys_execve() executes a new program.
1444  */
1445 static int do_execveat_common(int fd, struct filename *filename,
1446                               struct user_arg_ptr argv,
1447                               struct user_arg_ptr envp,
1448                               int flags)
1449 {
1450         char *pathbuf = NULL;
1451         struct linux_binprm *bprm;
1452         struct file *file;
1453         struct files_struct *displaced;
1454         int retval;
1455
1456         if (IS_ERR(filename))
1457                 return PTR_ERR(filename);
1458
1459         /*
1460          * We move the actual failure in case of RLIMIT_NPROC excess from
1461          * set*uid() to execve() because too many poorly written programs
1462          * don't check setuid() return code.  Here we additionally recheck
1463          * whether NPROC limit is still exceeded.
1464          */
1465         if ((current->flags & PF_NPROC_EXCEEDED) &&
1466             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1467                 retval = -EAGAIN;
1468                 goto out_ret;
1469         }
1470
1471         /* We're below the limit (still or again), so we don't want to make
1472          * further execve() calls fail. */
1473         current->flags &= ~PF_NPROC_EXCEEDED;
1474
1475         retval = unshare_files(&displaced);
1476         if (retval)
1477                 goto out_ret;
1478
1479         retval = -ENOMEM;
1480         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1481         if (!bprm)
1482                 goto out_files;
1483
1484         retval = prepare_bprm_creds(bprm);
1485         if (retval)
1486                 goto out_free;
1487
1488         check_unsafe_exec(bprm);
1489         current->in_execve = 1;
1490
1491         file = do_open_execat(fd, filename, flags);
1492         retval = PTR_ERR(file);
1493         if (IS_ERR(file))
1494                 goto out_unmark;
1495
1496         sched_exec();
1497
1498         bprm->file = file;
1499         if (fd == AT_FDCWD || filename->name[0] == '/') {
1500                 bprm->filename = filename->name;
1501         } else {
1502                 if (filename->name[0] == '\0')
1503                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1504                 else
1505                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1506                                             fd, filename->name);
1507                 if (!pathbuf) {
1508                         retval = -ENOMEM;
1509                         goto out_unmark;
1510                 }
1511                 /*
1512                  * Record that a name derived from an O_CLOEXEC fd will be
1513                  * inaccessible after exec. Relies on having exclusive access to
1514                  * current->files (due to unshare_files above).
1515                  */
1516                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1517                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1518                 bprm->filename = pathbuf;
1519         }
1520         bprm->interp = bprm->filename;
1521
1522         retval = bprm_mm_init(bprm);
1523         if (retval)
1524                 goto out_unmark;
1525
1526         bprm->argc = count(argv, MAX_ARG_STRINGS);
1527         if ((retval = bprm->argc) < 0)
1528                 goto out;
1529
1530         bprm->envc = count(envp, MAX_ARG_STRINGS);
1531         if ((retval = bprm->envc) < 0)
1532                 goto out;
1533
1534         retval = prepare_binprm(bprm);
1535         if (retval < 0)
1536                 goto out;
1537
1538         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1539         if (retval < 0)
1540                 goto out;
1541
1542         bprm->exec = bprm->p;
1543         retval = copy_strings(bprm->envc, envp, bprm);
1544         if (retval < 0)
1545                 goto out;
1546
1547         retval = copy_strings(bprm->argc, argv, bprm);
1548         if (retval < 0)
1549                 goto out;
1550
1551         retval = exec_binprm(bprm);
1552         if (retval < 0)
1553                 goto out;
1554
1555         /* execve succeeded */
1556         current->fs->in_exec = 0;
1557         current->in_execve = 0;
1558         acct_update_integrals(current);
1559         task_numa_free(current);
1560         free_bprm(bprm);
1561         kfree(pathbuf);
1562         putname(filename);
1563         if (displaced)
1564                 put_files_struct(displaced);
1565         return retval;
1566
1567 out:
1568         if (bprm->mm) {
1569                 acct_arg_size(bprm, 0);
1570                 mmput(bprm->mm);
1571         }
1572
1573 out_unmark:
1574         current->fs->in_exec = 0;
1575         current->in_execve = 0;
1576
1577 out_free:
1578         free_bprm(bprm);
1579         kfree(pathbuf);
1580
1581 out_files:
1582         if (displaced)
1583                 reset_files_struct(displaced);
1584 out_ret:
1585         putname(filename);
1586         return retval;
1587 }
1588
1589 int do_execve(struct filename *filename,
1590         const char __user *const __user *__argv,
1591         const char __user *const __user *__envp)
1592 {
1593         struct user_arg_ptr argv = { .ptr.native = __argv };
1594         struct user_arg_ptr envp = { .ptr.native = __envp };
1595         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1596 }
1597
1598 int do_execveat(int fd, struct filename *filename,
1599                 const char __user *const __user *__argv,
1600                 const char __user *const __user *__envp,
1601                 int flags)
1602 {
1603         struct user_arg_ptr argv = { .ptr.native = __argv };
1604         struct user_arg_ptr envp = { .ptr.native = __envp };
1605
1606         return do_execveat_common(fd, filename, argv, envp, flags);
1607 }
1608
1609 #ifdef CONFIG_COMPAT
1610 static int compat_do_execve(struct filename *filename,
1611         const compat_uptr_t __user *__argv,
1612         const compat_uptr_t __user *__envp)
1613 {
1614         struct user_arg_ptr argv = {
1615                 .is_compat = true,
1616                 .ptr.compat = __argv,
1617         };
1618         struct user_arg_ptr envp = {
1619                 .is_compat = true,
1620                 .ptr.compat = __envp,
1621         };
1622         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1623 }
1624
1625 static int compat_do_execveat(int fd, struct filename *filename,
1626                               const compat_uptr_t __user *__argv,
1627                               const compat_uptr_t __user *__envp,
1628                               int flags)
1629 {
1630         struct user_arg_ptr argv = {
1631                 .is_compat = true,
1632                 .ptr.compat = __argv,
1633         };
1634         struct user_arg_ptr envp = {
1635                 .is_compat = true,
1636                 .ptr.compat = __envp,
1637         };
1638         return do_execveat_common(fd, filename, argv, envp, flags);
1639 }
1640 #endif
1641
1642 void set_binfmt(struct linux_binfmt *new)
1643 {
1644         struct mm_struct *mm = current->mm;
1645
1646         if (mm->binfmt)
1647                 module_put(mm->binfmt->module);
1648
1649         mm->binfmt = new;
1650         if (new)
1651                 __module_get(new->module);
1652 }
1653 EXPORT_SYMBOL(set_binfmt);
1654
1655 /*
1656  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1657  */
1658 void set_dumpable(struct mm_struct *mm, int value)
1659 {
1660         unsigned long old, new;
1661
1662         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1663                 return;
1664
1665         do {
1666                 old = ACCESS_ONCE(mm->flags);
1667                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1668         } while (cmpxchg(&mm->flags, old, new) != old);
1669 }
1670
1671 SYSCALL_DEFINE3(execve,
1672                 const char __user *, filename,
1673                 const char __user *const __user *, argv,
1674                 const char __user *const __user *, envp)
1675 {
1676         return do_execve(getname(filename), argv, envp);
1677 }
1678
1679 SYSCALL_DEFINE5(execveat,
1680                 int, fd, const char __user *, filename,
1681                 const char __user *const __user *, argv,
1682                 const char __user *const __user *, envp,
1683                 int, flags)
1684 {
1685         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1686
1687         return do_execveat(fd,
1688                            getname_flags(filename, lookup_flags, NULL),
1689                            argv, envp, flags);
1690 }
1691
1692 #ifdef CONFIG_COMPAT
1693 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1694         const compat_uptr_t __user *, argv,
1695         const compat_uptr_t __user *, envp)
1696 {
1697         return compat_do_execve(getname(filename), argv, envp);
1698 }
1699
1700 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1701                        const char __user *, filename,
1702                        const compat_uptr_t __user *, argv,
1703                        const compat_uptr_t __user *, envp,
1704                        int,  flags)
1705 {
1706         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1707
1708         return compat_do_execveat(fd,
1709                                   getname_flags(filename, lookup_flags, NULL),
1710                                   argv, envp, flags);
1711 }
1712 #endif